wtf is paper-radar?

paper-radar is a personal academic literature tracking system, originally built by a physical medicine and rehabilitation doctor to keep up with new research papers without drowning in RSS feeds. It pulls updates from dozens of journal feeds and PubMed searches, scores each new paper against the user's own interest model, and pushes the results to a private web page for the user to review and mark on their phone or computer. The system runs as a scheduled job on a small always-on server. A fetch and score script collects papers from journal RSS feeds and PubMed searches, deduplicates them by DOI or title in a SQLite database, and ranks them using a keyword and topic weighting model so the highest-interest new papers rise to the top. A separate enrichment step checks each paper's DOI against Unpaywall to see if a free, open-access copy exists, and can optionally check institutional subscription access through a library link resolver, though the README is careful to note that actually downloading a paper still has to follow the user's institution and publisher terms. The private web page itself sits behind Cloudflare Access so only the owner can log in, with no public page or feed. From there, the user marks each paper as read, worth a quality review, worth summarizing, or gives it a thumbs up, neutral, or thumbs down rating, and these actions are stored in Cloudflare's D1 database so they sync across devices. A second version of the system, described as a personal research preference model, replaces the simple keyword scoring with a self-learning model that retrains nightly from all the user's past ratings, uses a statistical sampling method to balance showing familiar topics against occasionally surfacing unfamiliar ones, and includes a dashboard showing what topics the model thinks the user prefers. The project explicitly describes itself as the discovery stage of a three-part pipeline: paper-radar finds and ranks papers, a separate paper-fetch project downloads the full text, and another separate project called paper-review-and-digest handles quality appraisal and summarizing. The author notes the core radar can be self-hosted and reused, while the step that feeds results into personal notes is tied to their own Obsidian setup and offered only as an example to adapt.